Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Feature] Support ViTPose #1876

Merged
merged 14 commits into from
Mar 14, 2023
Merged

[Feature] Support ViTPose #1876

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
4b4e63c
Add ViTPose implementation on MMPose 1.x
LareinaM Dec 7, 2022
2601cd4
add pretrained config for backbone
LareinaM Dec 10, 2022
ce9b811
check for existence of attribute before access
LareinaM Dec 12, 2022
b424af8
fix formats
LareinaM Dec 12, 2022
d6807fd
fix formats
LareinaM Dec 12, 2022
4befb89
fix indentation and import order
LareinaM Dec 12, 2022
73bc29b
rename and add algorithm description
LareinaM Dec 23, 2022
8f0e873
follow changes in original repo
LareinaM Dec 27, 2022
4d6102c
Correct structure for simple decoders
LareinaM Feb 1, 2023
c46ffc6
Change configs, add val results, rename folder
LareinaM Feb 15, 2023
9f43c20
Fix formats
LareinaM Feb 15, 2023
085477e
Update markdown file
LareinaM Feb 17, 2023
c611933
Update training results
LareinaM Feb 23, 2023
23b7838
Update markdown file
LareinaM Mar 10, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
151 changes: 151 additions & 0 deletions ...ody_2d_keypoint/topdown_heatmap/coco/td-hm_ViTPose-base-simple_8xb64-210e_coco-256x192.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,151 @@
_base_ = ['../../../_base_/default_runtime.py']

# runtime
train_cfg = dict(max_epochs=210, val_interval=10)

# optimizer
custom_imports = dict(
imports=['mmpose.engine.optim_wrappers.layer_decay_optim_wrapper'],
allow_failed_imports=False)

optim_wrapper = dict(
optimizer=dict(
type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1),
paramwise_cfg=dict(
num_layers=12,
layer_decay_rate=0.75,
custom_keys={
'bias': dict(decay_multi=0.0),
'pos_embed': dict(decay_mult=0.0),
'relative_position_bias_table': dict(decay_mult=0.0),
'norm': dict(decay_mult=0.0),
},
),
constructor='LayerDecayOptimWrapperConstructor',
clip_grad=dict(max_norm=1., norm_type=2),
)

# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=210,
milestones=[170, 200],
gamma=0.1,
by_epoch=True)
]

# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)

# hooks
default_hooks = dict(
checkpoint=dict(save_best='coco/AP', rule='greater', max_keep_ckpts=1))

# codec settings
codec = dict(
type='UDPHeatmap', input_size=(192, 256), heatmap_size=(48, 64), sigma=2)

# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='mmcls.VisionTransformer',
arch='base',
img_size=(256, 192),
patch_size=16,
qkv_bias=True,
drop_path_rate=0.3,
with_cls_token=False,
output_cls_token=False,
patch_cfg=dict(padding=2),
init_cfg=dict(
type='Pretrained',
checkpoint='pretrained/mae_pretrain_vit_base.pth'),
),
head=dict(
type='HeatmapHead',
in_channels=768,
out_channels=17,
deconv_out_channels=[],
deconv_kernel_sizes=[],
loss=dict(type='KeypointMSELoss', use_target_weight=True),
decoder=codec,
extra=dict(upsample=4, final_conv_kernel=3),
),
test_cfg=dict(
flip_test=True,
flip_mode='heatmap',
shift_heatmap=False,
))

# base dataset settings
data_root = 'data/coco/'
dataset_type = 'CocoDataset'
data_mode = 'topdown'

# pipelines
train_pipeline = [
dict(type='LoadImage', file_client_args={{_base_.file_client_args}}),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size'], use_udp=True),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage', file_client_args={{_base_.file_client_args}}),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size'], use_udp=True),
dict(type='PackPoseInputs')
]

# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=4,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=4,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader

# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
149 changes: 149 additions & 0 deletions configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_ViTPose-base_8xb64-210e_coco-256x192.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,149 @@
_base_ = ['../../../_base_/default_runtime.py']

# runtime
train_cfg = dict(max_epochs=210, val_interval=10)

# optimizer
custom_imports = dict(
imports=['mmpose.engine.optim_wrappers.layer_decay_optim_wrapper'],
allow_failed_imports=False)

optim_wrapper = dict(
optimizer=dict(
type='AdamW', lr=5e-4, betas=(0.9, 0.999), weight_decay=0.1),
paramwise_cfg=dict(
num_layers=12,
layer_decay_rate=0.75,
custom_keys={
'bias': dict(decay_multi=0.0),
'pos_embed': dict(decay_mult=0.0),
'relative_position_bias_table': dict(decay_mult=0.0),
'norm': dict(decay_mult=0.0),
},
),
constructor='LayerDecayOptimWrapperConstructor',
clip_grad=dict(max_norm=1., norm_type=2),
)

# learning policy
param_scheduler = [
dict(
type='LinearLR', begin=0, end=500, start_factor=0.001,
by_epoch=False), # warm-up
dict(
type='MultiStepLR',
begin=0,
end=210,
milestones=[170, 200],
gamma=0.1,
by_epoch=True)
]

# automatically scaling LR based on the actual training batch size
auto_scale_lr = dict(base_batch_size=512)

# hooks
default_hooks = dict(
checkpoint=dict(save_best='coco/AP', rule='greater', max_keep_ckpts=1))

# codec settings
codec = dict(
type='UDPHeatmap', input_size=(192, 256), heatmap_size=(48, 64), sigma=2)

# model settings
model = dict(
type='TopdownPoseEstimator',
data_preprocessor=dict(
type='PoseDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
bgr_to_rgb=True),
backbone=dict(
type='mmcls.VisionTransformer',
arch='base',
img_size=(256, 192),
patch_size=16,
qkv_bias=True,
drop_path_rate=0.3,
with_cls_token=False,
output_cls_token=False,
patch_cfg=dict(padding=2),
init_cfg=dict(
type='Pretrained',
checkpoint='pretrained/mae_pretrain_vit_base.pth'),
),
head=dict(
type='HeatmapHead',
in_channels=768,
out_channels=17,
deconv_out_channels=(256, 256),
deconv_kernel_sizes=(4, 4),
loss=dict(type='KeypointMSELoss', use_target_weight=True),
decoder=codec),
test_cfg=dict(
flip_test=True,
flip_mode='heatmap',
shift_heatmap=False,
))

# base dataset settings
data_root = 'data/coco/'
dataset_type = 'CocoDataset'
data_mode = 'topdown'

# pipelines
train_pipeline = [
dict(type='LoadImage', file_client_args={{_base_.file_client_args}}),
dict(type='GetBBoxCenterScale'),
dict(type='RandomFlip', direction='horizontal'),
dict(type='RandomHalfBody'),
dict(type='RandomBBoxTransform'),
dict(type='TopdownAffine', input_size=codec['input_size'], use_udp=True),
dict(type='GenerateTarget', encoder=codec),
dict(type='PackPoseInputs')
]
val_pipeline = [
dict(type='LoadImage', file_client_args={{_base_.file_client_args}}),
dict(type='GetBBoxCenterScale'),
dict(type='TopdownAffine', input_size=codec['input_size'], use_udp=True),
dict(type='PackPoseInputs')
]

# data loaders
train_dataloader = dict(
batch_size=64,
num_workers=4,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_train2017.json',
data_prefix=dict(img='train2017/'),
pipeline=train_pipeline,
))
val_dataloader = dict(
batch_size=32,
num_workers=4,
persistent_workers=True,
drop_last=False,
sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
dataset=dict(
type=dataset_type,
data_root=data_root,
data_mode=data_mode,
ann_file='annotations/person_keypoints_val2017.json',
bbox_file='data/coco/person_detection_results/'
'COCO_val2017_detections_AP_H_56_person.json',
data_prefix=dict(img='val2017/'),
test_mode=True,
pipeline=val_pipeline,
))
test_dataloader = val_dataloader

# evaluators
val_evaluator = dict(
type='CocoMetric',
ann_file=data_root + 'annotations/person_keypoints_val2017.json')
test_evaluator = val_evaluator
Loading